 marfrit
|
0bd4222a36
|
iter1 phase6: implementation (file 3 follow-ups, close ka#6, write close artifact)
Phase 6 atoms executed:
1. Refined C4 to per-codec SSIM floors — documented in iter1_close.md.
2. Filed kernel-agent #11 [ka:experiment]: HEVC rkvdec_hevc_prepare_hw_st_rps
__pi_memcmp fault, m2m wedge. Includes dmesg trace + reproducer +
fresnel cross-check (RK3399 doesn't trip).
3. Filed kernel-agent #12 [ka:experiment]: VP9 enablement on RK3588
rkvdec via VDPU381/383 variant_ops. References memory
feedback_rkvdec_patch_reachability for path boundary.
4. Filed libva-v4l2-request-fourier #2: iter39 third-fd auto-probe
for RK3588 av1-vpu-dec. Pure backend, no kernel.
5. Updated kernel-agent #6 with per-iter1 outcome comment, closed
with cross-refs to #11, #12, libva-v4l2-request-fourier#2.
iter1_close.md applies the two Phase 5 framing amendments:
- H.264 SSIM 0.6676 reframed as 'independent RK3588 empirical
observation; converges with fresnel 0.6431 but neither is the
other's reference anchor'
- C7 reframed as 'out of iter1 scope (headless by design)' not
'rig blocker'
Carries-forward / does-not-carry state explicit for iter2/3/4 entry.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
|
2026-05-16 07:27:37 +00:00 |
|